Chemical Information Based Scaling of Molecular Descriptors:  A Universal Chemical Scale for Library Design and Analysis

Abstract

Scaling is a difficult issue for any analysis of chemical properties or molecular topology when disparate descriptors are involved. To compare properties across different data sets, a common scale must be defined. Using several publicly available databases (ACD, CMC, MDDR, and NCI) as a basis, we propose to define chemically meaningful scales for a number of molecular properties and topology descriptors. These chemically derived scaling functions have several advantages. First, it is possible to define chemically relevant scales, greatly simplifying similarity and diversity analyses across data sets. Second, this approach provides a convenient method for setting descriptor boundaries that define chemically reasonable topology spaces. For example, descriptors can be scaled so that compounds with little potential for biological activity, bioavailability, or other drug-like characteristics are easily identified as outliers. We have compiled scaling values for 314 molecular descriptors. In addition the 10th and 90th percentile values for each descriptor have been calculated for use in outlier filtering.

Molecular Property Filters Describing Pharmacokinetics and Drug Binding

“…Such molecular properties were also adopted for the prediction of complex physiological properties and pharmacokinetics: absorption [12], or blood brain barrier penetration [13,14], and their use culminated in the definition of general drug-likeness ranges. These empirical ranges of the properties were proved to be useful as property filters in the design of compound libraries of drug screening [15][16][17][18][19][20]. Notably, the selection of high quality (drug-like) compound libraries [3,[21][22][23]] is a primary and key step of the screening process.…”

Examination of the Computed Molecular Properties of Compounds Selected for Clinical Development

“…Adopted Names [USAN] or International Nonproprietary Names [INN] names extracted from the World Drug Index) with compounds not presumed to be drug-like (those found in the Available Chemical Directory [ACD]) and determined that compounds with excessive logP, molecular weight (MW), hydrogen-bond donors, or acceptors were more likely to have solubility and/or permeability problems that would lead to poor oral bioavailability (6). Since the publication of the rule-of-five, there have been many additional studies aimed at distinguishing between drugs and nondrugs (7)(8)(9), characterizations of the properties of drug-like databases (10)(11)(12)(13), and the prediction of oral bioavailability (14)(15)(16). From these exercises, one would hope to define the appropriate molecular properties from which leads or compounds targeted for clinical investigation should be derived.…”

A Cheminformatic Toolkit for Mining Biomedical Knowledge

“…Cheminformatics research centers will perform investigations whose results can be applied towards the (1) analysis, visualization, and interpretation of screening data (32,33); (2) efficient planning of new syntheses and assays (32,(34)(35)(36)(37)(38)(39); and (3) rapid advancement of screening results towards therapeutically relevant applications (40)(41)(42)(43)(44)(45)(46)(47). Currently, the National Center of Biotechnology Information (NCBI) at NIH is developing the PubChem database of chemical structures and biological activities (48).…”